3D CITY MODELLING OF ISTANBUL HISTORIC PENINSULA BY COMBINATION OF
AERIAL IMAGES AND TERRESTRIAL LASER SCANNING DATA
S. Buhura, Th. Kerstenb, G. Büyüksaliha, K. Jacobsenc, I. Baza, S. Dursuna, D. Sagira
a
Bimtas, Tophanelioglu Cad. ISKI Hizmet Binasi No:62 K.3-4, 34460 Altunizade-Istanbul, Turkey - ibaz@gyte.edu.tr,
gbuyuksalih@yahoo.com, selvidursun@hotmail.com, dileksagir_@hotmail.com
b
HafenCity University Hamburg, Department Geomatics, Hebebrandstrasse 1, D-22297 Hamburg, Germany
- thomas.kersten@hcu-hamburg.de
c
Leibniz University Hannover, Institute of Photogrammetry and Geoinformation, Nienburger Str. 1, D-30167
Hannover, Germany - jacobsen@ipi.uni-hannover.de
Commission VII, WG 6
KEY WORDS: Aerial, City, Cultural Heritage, Integration, Modelling, Project, TLS, Visualization
ABSTRACT:
There is an increasing demand for 3D city models for many applications and users worldwide. Some of this growth in demand has
been caused by the increase in public availability of open geospatial viewers (e.g. Google Earth, Virtual Earth). Although the
Historic Peninsula of old Istanbul was added to the UNESCO World Heritage List in 1985, no documentation of this important area
has yet been carried out. In 2006 the Istanbul Greater Municipality’s Directory of the Protection of Historical Environment initiated
the “Historical Peninsula project”, which comprises a project area of 1500 ha and approximately 48,000 buildings in crowded and
frequently narrow streets. Therefore, BIMTAS, a company of the Greater Municipality of Istanbul, started the documentation and
modelling of all buildings in the area of the Historic Peninsula by terrestrial laser scanning (TLS) and aerial imagery in the same
year. Although the data acquisition/recording of the project area could be completed within a time slot of 15 months, the 3D
mapping and modelling turned out to be a hugely time-consuming challenge. The entire production environment from data
acquisition to 3D mapping and modelling of the buildings is described in this paper, whereby the focus is emphasized on the latter.
The combination of terrestrial laser scanning data and aerial imagery for deriving new products of 3D city models at different levels
of detail is presented.
1. INTRODUCTION
Several disciplines like urban planning, architecture,
telecommunication, tourism, environmental protection and
many others have an increasing demand for digital 3D city
models, in order to use such complex data for planning,
analyses, visualization and simulation in different applications.
Additionally, the open geospatial viewers (e.g. Google Earth,
Virtual Earth) increase the demand on 3D city models. To
satisfy this increasing demand for such data, the city models
must be acquired quickly, precisely, in detail, and with full
completeness and in an economic manner. The Historic
Peninsula of old Istanbul (Fig. 1) is one of the most important
tourism locations in Turkey and is a challenge for 3D city
modelling due to its complex building and roof structures.
and 3D modelling with the combination of both data is
described.
Due to its importance these "Historic Areas of Istanbul" were
added to the UNESCO World Heritage List in 1985. It is
located on the southern shore of the Golden Horn, which
separates the old city centre from the northern and younger
parts of the European side. The Historic Peninsula ends with the
Theodosian land walls in the west. The peninsula is surrounded
by the Sea of Marmara on the south and the Bosporus on the
east.
A brief introduction into the Historic Peninsula project is given.
The production environment for roof mapping using aerial
imagery, façade mapping with terrestrial laser scanning data
Figure 1. Istanbul Historic Peninsula
2. THE ISTANBUL HISTORIC PENINSULA PROJECT
The inner city wall area of Istanbul known as Historical
Peninsula consists mostly of archaeological, urban and
1239
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
historical protected areas. The Historic Peninsula (Fig. 1)
comprises an area of 1500ha with approximately 48,000
buildings in crowded and frequently narrow streets (total length
400 km). The facades of the building along the roads and streets
cover an area of about 5,500 000 m². In 2003 it was completely
declared as a protected area, when urban protection plans at
1:5000 and 1:1000 map scales were completed. For these
protected areas detailed studies of urban design projects based
on 1:500 and 1:200 map scales shall be carried out in the future.
calibration of the system in the streets of Istanbul took some
weeks, but the data acquisition in the field has been working
since the end of June 2007. The laser scanner was fixed with its
orientation in the horizontal direction, scanning only in the
profile perpendicular to the moving direction and operating
with a speed of up to 40 profiles/sec. The distance between
neighbouring profiles was 2-3 cm at the beginning,
corresponding to a speed of the van during scanning of
0.5m/sec up to 0.75m/sec or 1.8 km/h up to 2.7km/h.
A project contract, briefly named as “Historical Peninsula
project” was allocated to BİMTAŞ by the Istanbul Greater
Municipality’s Directory of the Protection of Historical
Environment. Finally, due to the request of many municipality
applications and due to an expected earthquake in Istanbul
within the next 30 years BIMTAS, a company of the Greater
Municipality of Istanbul, started the documentation of all
buildings in the Historic Peninsula area by terrestrial laser
scanning in 2006. It has been planned that the Historic
Peninsula should be mapped in a time frame of two years,
which demonstrates the great ambition of the project.
The data acquisition of the Historic peninsula was carried out
by terrestrial laser scanning for the building facades in the
streets and for the building roofs by aerial photo flights with an
analogue and a digital camera.
3. DATA ACQUISITION BY TERRESTRIAL LASER
SCANNING
The data acquisition by terrestrial laser scanning started in
September 2006 mainly using four Leica HDS4500. Figure 2
shows an example of a coloured point cloud of building facades
in the Historic Peninsula. 80ha of the project area (of 1500ha in
total) was scanned within the first six months using the existing
production capacity, which clearly indicated, that the scanning
would need more than eight years for the entire project area, if
this current scan rate of approximately 0.7ha per day could not
be increased.
Figure 3. Sensor configuration on the mobile mapping van of
VISIMIND AB
Thus, the speed of data acquisition by terrestrial laser scanning
was significantly increased through use of this mobile mapping
system. Consequently, the laser scanning with the mobile
system was finished by November, 8th, 2007 with the improved
total production rate of ~600m per hour, while post processing
of the multiple sensor data took until January 2008. The
production rate was mainly 1:10, i.e. for one hour scanning 10
hours post processing of the data was needed. However,
approximately 2% of the area (30ha) could not be scanned by
mobile TLS due to traffic restrictions and environmental
conditions. For the scanning of this remaining part static TLS is
required. A more detailed description of the data acquisition by
terrestrial laser scanning for the Historic Peninsula project is
summarised in Baz et al.,(2008).
4. MAPPING OF FACADES
The geo-referenced point clouds from laser scanning were used
for line mapping of the facades at a scale of 1:200. The required
positional standard deviation of 0.2mm on the map corresponds
to 4cm in the object space as relative accuracy. The facade
mapping was carried out by 34 operators using the Mencisoftware Z-MAP Laser from Italy, which is able to process laser
scan data and rectified photogrammetric images simultaneously
for line mapping with limited AutoCAD functionality. An
example for the mapping of building facades with Z-MAP
Laser is shown in Fig. 4.
Figure 2. Coloured point cloud of building facades in the
Historic Peninsula
As a consequence the static scanning was replaced by mobile
mapping through the Swedish company VISIMIND AB (Fig. 3)
in June 2007 using a hybrid sensor system on the vehicle
consisting of a terrestrial laser scanning system HDS4500,
supported by GPS/IMU and digital cameras. This increased the
scan rate dramatically. The sensor integration and the
1240
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
Figure 6. Detailed map of building façades using laser scanning
data - laser scanning data (top) and façade plan (bottom)
Figure 4. 3D polylines map of building facades based on ZMAP Laser
The production rate varied between 60 m2 facade/day/operator
(March 2007) and 140 m2/day/operator (October 2007), which
is an increase of production speed by a factor of more than 2. If
one assumes in total 5 Mio m2 façade areas for mapping of the
Historic Peninsula, it corresponds to an estimated mapping time
of approximately five years with 34 operators working on 210
days per year.
An example of the final product from façade mapping is
depicted in figure 6, which is derived from 3D polylines as
illustrated in figure 5. The mapping of the facades is described
in more detail in Baz et al.,(2008).
Figure 5. Mapped 3D polylines of building block façades
derived from terrestrial laser scanning data
5. ROOF MAPPING WITH AERIAL IMAGERY
Since early July 2007 a roof mapping group was established in
order to measure and to model the roofs of all buildings in 3D
within the Historic Peninsula project. A project team of five
operators started the new production line after three days of
intensive training in mid July using the Z-MAP Foto software.
In the beginning available UltraCamD images with 30cm
ground sample distance (GSD) were used for data acquisition.
The synthetic UltraCamD image has 7500 x 11500 pixels with a
pixel size of 9µm*9µm corresponding to an image format of
67.5mm in the flight direction and 103.5mm across flight
direction and a focal length of 101.4mm. Each image covered
2.3km x 3.5km.
The photo coordinates have been determined by automatic
aerotriangulation with an Intergraph workstation, while the
orientation elements of the digital images were adjusted by
bundle block adjustment with BLUH. The results of the bundle
block adjustment with self-calibration of the UltraCamD data
are summarised in (Büyüksalih and Jacobsen,2006). A
horizontal accuracy of up to a factor 0.6 of the GSD could be
achieved at independent check points, which corresponds to
18cm in object space. This is a limited result for a digital
camera. The sigma0 from the bundle block adjustment is in the
range of 0.4 pixels, indicating that the accuracy potential is
better. The main reason for this is the limited control point
definition and accuracy. Nevertheless the achieved geometric
quality is sufficient for the roof mapping, which is more limited
by the object definition. The vertical accuracy is two times
better than the horizontal accuracy due to good connections
between the images in the block and due to measurements of
each control point in 4.5 images (on average).
However, due to the limited resolution of the UltraCamD
imagery it was very difficult for the operators to measure small
roofs and to identify clearly the roof points. As a rule of thumb,
mapping is possible up to 0.1mm GSD in the map scale,
corresponding to a map scale of 1:3000 with 30cm GSD, which
was confirmed by the tests made by this group. Thus, it was
decided to use higher resolution imagery for this task, available
since mid of August as scanned analogue colour aerial images
with 9.5cm GSD. The aerial flight has been conducted using a
JenOptik LC0030 camera (f = 305mm) in July 2006 at a photo
scale of 1:4500 covering almost 1 km2 per photo. The photo
orientation was determined by automatic aerial triangulation
and bundle block adjustment. The expected standard deviation
is Sxy = 10cm and Sz = 14cm for topographic points, which is
much better than from the triangulation of the UltraCamD
images. The analogue photos were scanned at a resolution of
21µm using a Zeiss SCAI scanner. As an example a part of an
1241
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
aerial image taken by the analogue camera illustrates the
configuration of roofs in the Historic Peninsula (Fig. 7).
Figure 7. Part of an aerial image from the Historic Peninsula
used for roof mapping
The advantage of the analogue imagery is the high resolution of
approximately 10cm GSD in comparison to the limited 30cm
GSD of the UltraCamD images, which enables clear
identification of the roof points by the operator. In contrast,
measurement in shadow areas of the scanned analogue images
is not possible due to limited dynamic range of the photos. In
comparison to images from digital cameras the contrast and
brightness adjustment in analogue images does not provide
satisfying results in shadow areas for measurement tasks.
In this part of the project to ensure the quality of the data it was
an essential task to combine the two different data sets from
aerial imagery and mobile TLS into one common data set in the
same coordinate system without any discrepancies from the
different data acquisition sources. Therefore, the orientation of
the aerial images has been transformed to the same datum as
used for the laser scanning data, in order to perform the
mapping in the same coordinate system. The differences
between façade corners and roof corners are mainly in the range
of 20-60cm (spatial vector). The differences represent mainly
the effect of point definition – the roof extends over the wall. In
addition the following error sources exist: effects from datum
transformation, accuracy of the orientation data, accuracy of the
laser scanning data, identification of the roof corners in the
images, definition of the roof corner and facade corner (rain
spout), respectively. The two last sources for discrepancies
might have the biggest influence on the accuracy of the merged
data, which is generated from data of two different sensor
systems (terrestrial laser scanning and aerial images). Two
examples of the combination of 3D polylines of roofs from
aerial images and 3D polylines of facades from terrestrial laser
scanning data are presented in Fig. 8.
The production rate is approximately 100ha of mapped rooftops
within 30 days (status August 2007) with five operators. It is
estimated that the area of the Historic Peninsula project (1500
ha) will be mapped within one year.
Figure 8. Combination of data from two different sources: 3D
polylines of roofs based on aerial images and 3D polylines of
facades from terrestrial laser scanning data
6. 3D MODELLING OF CITY MODELS AND
LANDMARKS
Digital photogrammetry and laser scanning, both airborne and
terrestrial, are efficient modern techniques for 3D data
acquisition as a base for 3D modelling. Image-based 3D
modelling has been reviewed by Remondino and El-Hakim
(2006). The 3D modelling for specific products using different
data sources is the most time-consuming task. Therefore, it was
decided to generate different products with different level of
detail (LoD) and quality for the city model and the landmarks.
The definition of the level of detail is taken from the 3D city
model of Hamburg: LoD1 = block model as an extruded foot
print of vector data, LoD2 = roof model with detailed roof
structures, and LoD3 = architectural model with detailed roofs
and facades. LoD4 would include a detailed exterior and
interior model. For 3D modelling of all products Autodesk®
3ds Max® software was used. This software is a powerful,
integrated 3D modelling, animation, and rendering application
to efficiently create parametric shapes and objects and to
quickly begin modelling. 3ds Max offers also various
import/export functions for several formats, which provides a
huge flexibility for the requirements of the clients.
6.1 3D city model in level of detail 1 (LoD1)
In December 2007 BIMTAS started to generate a 3D city model
of the Historic Peninsula as a so-called block model (LoD1)
using just cadastre data. This model was completed for all
50,000 buildings after only three months of production with two
operators. LoD1 product (Fig. 9) has been generated from the
building outlines available in the vector map of the Historic
Peninsula, while the building heights were extracted from
known information about the number of storeys above the
horizontal surface plane. Currently the data is available in 3dsformat in 1km²- tiles.
1242
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
Figure 10. 3D city model of Istanbul Historic Peninsula as
LoD2 in the Suleymaniye area superimposed on an orthophoto
as an overview (top) and detailed view (bottom)
6.3 3D city model in level of detail 3 (LoD3)
Figure 9. 3D city model of Istanbul Historic Peninsula as LoD1
6.2 3D city model in level of detail 2 (LoD2)
Currently, the modelling group is producing a 3D city model in
LoD2 (Fig. 10) by combining the block model with the roofs,
which are measured as 3D polylines by the roof mapping group.
Therefore, the measured polylines of the roofs will be converted
to 3D planes and thereafter the roof corners can be connected
perpendicularly to the corresponding roof points of the block
model. In this modelling process the building heights of each
block model will be matched with the height of the measured
roof. It was quickly realised that there were some discrepancies
between the block model and the roof model, i.e. the vector data
from cadastre did not fit very well with the
photogrammetrically determined roofs. Therefore, it was
decided to change the method of production by using measured
height points on the ground to generate the facades, i.e. the roof
corners can be connected perpendicularly to the measured
ground points.
Currently, the Suleymaniye region could be almost finished as
LoD2 product as illustrated in Fig. 10.
The LoD3 is a high quality product, which is generated from
the detailed roof-mapping and from building facades including
photo-realistic texture mapping of the each building. Similar
detailed models of north German castles were generated by
digital photogrammetry (Kersten et al., 2004) or archaeological
3D objects in the Republic of Yemen by a combination of
photogrammetry and terrestrial laser scanning (Kersten, 2007).
Due to the very time-consuming generation of such products
only one project part, as pilot project, has been completed thus
far. For an architectural comparison of an existing and planned
situation two 3D models (old/new) of the same block (511) in
Suleymaniye Bölgesi were generated. The modelling of the
planned situation was based on 2D drawings, while the existing
situation could be modelled using the facade drawing data,
which was generated from laser scanning data. For the existing
buildings photo realistic textures were available, while for the
planned situation textures from the library of ArchiCAD were
used. Following a request from the project architects a video
sequence (4:47min, 640 x 480 pixel, 46 MB as wmv format)
was generated for a visual comparison of the existing and the
planned situation of a block in Suleymaniye Bölgesi.
6.4 3D modelling of landmarks
The 3D modelling group generated photo-realistic 3D models
of landmarks, such as the “German Fountain” shown in figure
11, with data from façade mapping using 3ds max. Additionally,
the group was able to perform 3D modelling using existing
ground and façade plans from different sources (e.g. Sultan
Ahmet Mosque, see Fig. 12). The details and the precision of
these models are more than sufficient for visualisation purposes.
Finally, video sequences can be generated from any modelled
object using 3ds max.
1243
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
Figure 11. Photo-realistic and detailed 3D model of the
“German Fountain” (3ds max)
Figure 13. Photo-realistic and detailed 3D model of a historic
grave located in the garden of the Suleymaniye Mosque
Figure 12. Perspective Scene of the “German Fountain” and the
Sultan Ahmet Mosque (3ds max)
Thus far the following landmarks have been modelled in 3D:
German Fountain and the Sultan Ahmet Mosque (Blue Mosque),
a historic grave located in the garden of the Suleymaniye
Mosque (Fig. 13), and the fountain of the Yildiz palace.
Currently, the 3D modelling is focussed on the Suleymaniye
Mosque (Fig. 14).
However, one major problem in modelling and visualisation is
still not solved sufficiently: The data volume of the CAD data
for the 3D models (e.g. 60 MB DWG file for the German
Fountain) is far too large for many visualisation applications
due to the high level of detail. This problem might be solved in
the future by better computer performance. For (interactive)
visualisation of the 3D models on the internet an export of
photo-realistic models from 3ds max to standard file formats
(e.g. VRML) is required. Hence, 3D models with reduced data
volumes for better visualisation performance on the internet
must be generated. Which is the most appropriate file formats
for visualisation of the 3D models and for data exchange? There
is still potential for the optimisation of the CAD data by using
geometric features instead of poly-meshes.
Figure 14. Wireframe of the Suleymaniye Mosque 3D model
7. CONCLUSION AND OUTLOOK
Since 2006 BIMTAS has been tackling a large project on the
architectural documentation of Istanbul’s Historic Peninsula
including almost 50,000 buildings (11,000 of these buildings
are historic) with an area of 1500ha. In the frame of this project,
the company established different production groups (e.g. laser
scanning and geodesy group, and groups for façade mapping,
roof mapping and 3D modelling) with highly educated and
trained staff (geodesists and architects), who were able to tune
and to optimise the production lines for efficient cultural
heritage documentation. In order to run such a huge project,
modern technologies (hardware and software) were required.
With the Historic Peninsula project BIMTAS was able to build
up a modern production environment with high-end technology
and sophisticated personnel, which could efficiently perform
3D mapping of building facades and roofs as a data base for the
subsequent 3D modelling. The data acquisition of all the
building facades in the streets of the Historic Peninsula was
finished in less than three months by mobile terrestrial laser
scanning. However, processing and 3D mapping of the laser
scanning data are still continuing. A similar project is reported
by Frueh and Zahkor,(2003), who also used ground-based data
1244
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
from mobile laser scanning system and cameras in addition to
airborne laser scanning data and aerial imagery.
With 3D city models in LOD1, LOD2 and LOD3 BIMTAS is
going to derive new products from laser scanning data and
aerial imagery, which will fulfil the requirements of many
applications and users due to the different levels of detail. The
dominating parts of the production lines for the generation of
such 3D city models are still manual work, i.e. there is huge
potential for optimisation and automation of the production to
complete such projects as the Historic Peninsula in reasonable
time frames. In the meantime the 3D mapping and modelling of
landmarks will be continued in order to generate the most
important landmarks as interactive textured 3D models
Istanbul is selected as one of European Cultural Capitals in
2010. Thus, many other projects will come in the near future.
There is already a request to create the LoD1 and LoD2
products of the area comprising the two sides of the lower part
of the Bosporus including approximately 15,000 buildings.
BIMTAS learned many lessons concerning project management
and tuning of technology to manage the requirements, so that
the company is well positioned to run similar projects in the
future.
8. REFERENCES
Baz, I., Kersten, Th., Büyüksalih, G., Jacobsen, K.,2008.
Documentation of Istanbul Historic Peninsula by Static and
Mobile Terrestrial Laser Scanning. International Archives of
Photogrammetry, Remote Sensing and Spatial Information
Sciences, Vol. XXXVII, ISPRS Congress, Beijing, July, 3-11.
Büyüksalih, G., Jacobsen, K.,2006. Bundle Block Adjustment
with UltraCamD Images.
International Archives of
Photogrammetry, Remote Sensing and Spatial Information
Sciences, Vol. XXXVI, Part I, ISPRS Commission I, Paris,
SFPT No. 184 (2006-4), pp. 17 – 22.
Frueh, C., Zakhor, A.,2003. Constructing 3D city models by
merging ground-based and airborne views. Proceedings of
IEEE Computer Society Conference on Computer Vision and
Pattern Recognition, Vol. 2, pp. 562-569.
Kersten, Th.,2007. Virtual Reality Model of the Northern
Sluice of the Ancient Dam in Marib/Yemen by Combination of
Digital Photogrammetry and Terrestrial Laser Scanning for
Archaeological Applications.
International Journal of
Architectural Computing, Special Focus on Cultural Heritage,
Issue 02, Volume 05, Published by Multiscience, pp. 339 - 354.
Kersten, Th., Acevedo Pardo, C., Lindstaedt, M.,2004. 3D
Acquisition, Modelling and Visualization of north German
Castles by Digital Architectural Photogrammetry. International
Archives of Photogrammetry, Remote Sensing and Spatial
Information Sciences, Vol. XXXV, Commission V, Part B2, pp.
126-132, presented paper at the XXth ISPRS Congress, Istanbul,
July, 2004.
Remondino, F., El-Hakim, S.,2006. Image-based 3D modelling:
a review. The Photogrammetric Record, Vol. 21 (115),
September, pp. 269-291.
1245
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
1246